Answers from the documentation; checking which model is on duty
Hi - I answer from the OpenSmartRoute documentation: routing, the API, plans and quotas, self-hosting. Ask away, or open a support ticket if you need a person.
Grounded in the docs - follow a source before acting on it.
Open Library of 2,000 Simulation-Ready 3D Assets for Robotics - OpenSmartRoute
Hugging Face launched SimReady Central. This open library holds over 2,000 3D assets. These assets are ready for robotics simulations. They also suit AI factory digital twins.
Building robotics simulations requires more than just 3D models. Developers often need extra properties like collision geometry. They also need mass and friction data. SimReady assets come prepared with these properties. This reduces the manual work for teams.
SimReady Central brings together OpenUSD assets. OpenUSD is an open framework for describing 3D scenes. These assets are free and ungated. Teams can use them for physical AI training. They can also use them for factory digital twins.
The library includes infrastructure and equipment. It lists robot arms, grippers, and warehouse racks. It also lists cooling systems. Developers can check which capabilities each asset supports. They can combine assets from different suppliers. This creates a complete scene for testing.
Equipment manufacturers can prepare assets for the library. They validate assets against SimReady requirements. This helps them reach developers quickly. End customers can test scenarios faster. Industries can build simulations around their products. They manage their own datasets on Hugging Face.
SimReady assets need specific properties for simulation. A detailed CAD model is a good start. Simulation requires information about physical behavior. It needs data on how objects interact.
For robot manipulation, mass and friction are key. AI factories need electrical or thermal metadata. OpenUSD scenes describe how objects compose. Training autonomous machines requires collision data. It also needs sensor interaction info.
The SimReady Foundation defines these requirements. It provides profiles and validation tools. A profile groups features for a use case. It lists what an asset must support.
Features include collision geometry and physics. They include articulated robot movement. They include object identification and synthetic data. They include rendering and sensor simulation. Developers select assets matching their workflow. They compose them into an OpenUSD scene.
OEMs publish assets to make them discoverable. They turn CAD models into SimReady assets. This helps teams building simulations find equipment. Manufacturers reach teams making digital twins.
Day 2 of TechCrunch Disrupt 2026 features sessions on robotics, agent security, and the shift from chatbots to physical AI systems. Speakers discuss funding, hardware, and the new rules for building AI in the enterprise.
Microsoft will scale its Maia accelerator family to multi-gigawatt deployments from 2027, making it the primary inference system for its own products. Maia 300 collapses scale-up and scale-out onto one Ethernet fabric and moves KV state ...
AI factory OEMs like Vertiv share specific assets. They provide warehouse equipment racks. They provide server racks and cooling gear. Developers reuse the geometry and properties. Frontier agents incorporate these assets into twins.
Robotics OEMs like Comau and FANUC share arms. They share grippers and end-effectors. Teams building training environments find these easily. Developers combine them with industrial equipment. They build warehouse and factory scenarios.
Palatial provides the NVIDIA Robotics 1K Dataset. This dataset includes complex articulated assets. These assets come from text, images, and scans. Developers build assets for robot training. They use them for evaluation tasks.
LG Electronics shares assets for home robots. They share refrigerators and dishwashers. Developers use assets informed by appliance experience. They build household simulation environments. They generate synthetic data for training.
ForgeCAD provides articulated robotics props. These props have high-quality textures. Developers use them for perception training. They use them for manipulation training. They generate synthetic data for testing.
OEMs use a submission skill to validate assets. They package assets for inclusion in the library. They submit them for review. This process ensures quality standards.
NVIDIA USD Search powers natural-language search. It helps developers find assets across datasets. Hugging Face integrated this into their platform search. It indexes USD properties and simulation metadata. Agents discover assets suited for physical tasks.
AI-assisted authoring helps build environments faster. Developers use frontier models like GPT-6 Astra. They combine asset search with scene-authoring tools. Astra finds suitable assets based on constraints. It writes code to assemble a scene. It creates scenario variations for testing.
A developer can prompt Astra to build a pick-and-place scenario. The prompt asks for a warehouse scenario in Isaac Sim. It requests a FANUC arm and a Robotiq gripper. It asks for Python code to assemble the scene. It asks for a .mov preview file.
Astra queries the SimReady Search API. It finds relevant assets on Hugging Face. It references assets into an OpenUSD stage. It generates Python code to assemble the workcell. This turns requirements into a reviewable simulation.
The same approach supports AI factory digital twins. The task shifts to equipment placement. Developers arrange racks and cooling equipment. They follow specified spatial constraints. They check for equipment overlaps. They report clearance violations.
Astra assembles selected SimReady assets into layouts. It generates scripts to check aisle clearances. Developers inspect layouts in their simulation app. They compare equipment placement and access paths. They review flagged issues before refining.
Developers can extend environments in several ways. They can reuse the environment with separate layers. They can vary scenarios with additional assets. They can build application-specific tests. They can carry validated layouts into operations.
Teams can connect scenes to live telemetry. This keeps the digital twin synchronized. It helps with power distribution analysis. It helps with cooling performance analysis.
The roadmap includes more OEM partners. The asset library is growing. Community feedback identifies needed additions. Teams tell the team which assets they need. They share which runtimes they use.
OEM contributions broaden available content. Developer examples show how assets work. Teams share examples of what they build. They open issues for bugs or feature requests.
Teams should choose an asset for their next project. They can find assets in the SimReady Central Library. They can search using natural language. They can build their own asset browser. They can learn to assemble OpenUSD scenes. They can develop robotics and digital twin skills.
They can start with Isaac Sim and Isaac Lab. They can build AI factories using the library.
For teams using OpenSmartRoute, this news offers clear routing benefits. OpenSmartRoute sends requests to the best-fit model or agent. It scores candidates on quality, cost, speed, and safety. The team sets weights per request. Hard rules pin requests to specific constraints.
A request with personal data stays on an on-premises model. A region or cost cap is never crossed. The router learns from outcomes. A model that answers well gets more traffic. A model that fails gets less traffic.
A new model is one catalogue entry. It competes on the next request. Nothing else changes in the app. OpenSmartRoute works with any OpenAI-compatible provider. It works with open-weight models served locally. It works with MCP tools and A2A agents.
The savings ledger shows what each routed request cost. It compares this to the most expensive model cost. This tracks efficiency for every request. The input guard spots prompt injection. It spots personal data before a request leaves.
The 'osr eval' tool measures routing accuracy. It uses the team's own prompts. It can fail a build when accuracy drops. The hosted platform keeps a models catalogue. It lists prices and public rankings. These rankings are built from real traffic.
Teams can set up a catalogue entry for SimReady assets. They can define rules for asset selection. They can set cost caps for simulation runs. They can configure safety checks for physical data.
Teams can monitor the savings ledger daily. They can track how much OpenSmartRoute saves. They can adjust weights based on simulation quality. They can ensure personal data stays on-premises.
Teams can use the input guard to scan prompts. They can verify no personal data leaks. They can ensure region compliance for every request. They can confirm cost caps are met.
Teams can run 'osr eval' before deploying changes. They can check routing accuracy on their prompts. They can fix issues if accuracy drops. They can ensure the router performs well.
Teams can update the catalogue with new SimReady assets. They can add new models for simulation tasks. They can test new models against the catalogue. They can ensure the router finds the right assets.
Teams can share their simulation scenarios with the community. They can contribute assets to SimReady Central. They can help grow the open library. They can improve the ecosystem for everyone.
Announcement of SimReady Central and its core purpose
NVIDIA announced SimReady Central on Hugging Face. It is an open library for robotics simulation. The library holds more than 2,000 simulation-ready assets. These assets are OpenUSD files. They include robot arms, grippers, and warehouse racks. They also include cooling infrastructure. Developers can use these files for free. No one pays a fee to access them. The goal is to speed up AI factory digital twins. Teams do not need to build models from scratch. They can pick assets from the library. This reduces the time to start simulation work.
What makes an asset SimReady and the required properties
An asset becomes SimReady after validation. It must meet specific technical requirements. These requirements come from the SimReady Foundation. The foundation defines clear profiles for use cases. A profile lists needed features like collision geometry. It also lists mass and friction properties. For AI factories, it includes electrical metadata. It includes thermal metadata too. SimReady assets replace raw CAD files in workflows. They come with physics data already attached. This data helps robots understand how objects move. It helps agents predict collisions. Developers check these properties before using an asset.
How OEMs contribute assets to the open library
OEMs prepare their equipment models for the library. They use the SimReady submission skill. This skill validates assets against profiles. OEMs package the validated files. They submit them to SimReady Central. This process makes assets discoverable. Manufacturers reach teams building simulations. End customers can test scenarios faster. OEMs manage their own Hugging Face datasets. They do not lose control of their data. The library acts as a bridge between makers and users. It allows industries to build simulations around products.
Specific examples of equipment from major industry partners
Vertiv provides warehouse equipment racks for AI factories. They provide server racks and cooling equipment. Their assets support digital twins for data centers. Comau, FANUC, and Franka Robotics offer robot arms. Robotiq and Standard Bots provide grippers. Universal Robots adds more robotic tools. Palatial released the NVIDIA Robotics 1K Dataset. This dataset includes complex articulated assets. The assets come from text, images, and scans. LG Electronics shares refrigerator and dishwasher models. These models support home robotics simulations. ForgeCAD offers articulated robotics props with textures. These props help with perception training. Each partner contributes unique equipment types.
How developers use assets to build robotic pick-and-place scenarios
Developers find assets in SimReady Central. They select assets matching their workflow needs. They combine assets into an OpenUSD scene. The scene includes robot arms and objects. Developers configure collision geometry in the scene. They set mass and friction values. This setup allows robots to grasp objects. It allows robots to move items safely. Developers can train agents on these scenes. They can evaluate robot performance easily. The scene supports sensor simulation too. Developers can test vision systems on the assets. This testing happens without physical hardware. It saves time and money on prototyping.
Using AI agents to assemble factory layouts and check clearances
AI agents can assemble factory layouts automatically. They use SimReady assets as building blocks. Agents check if paths are clear between robots. They ensure equipment does not block movement. Agents verify electrical connections in server racks. They check thermal spacing for cooling units. This process replaces manual layout drawing. It reduces human error in planning. Agents can simulate thousands of layout options. They pick the best configuration for the factory. Teams can test safety rules digitally. They can see collisions before building anything.
Why this changes how teams build robotics simulations and digital twins
This news changes the workflow for simulation teams. Teams no longer build every asset from scratch. They reuse verified simulation-ready assets. This speeds up the creation of digital twins. It lowers the barrier to entry for robotics AI. New teams can start with a library of tools. They do not need deep engineering knowledge. They can focus on training models instead. The open library grows as more people contribute. It creates a shared resource for the industry. Simulation becomes faster and more accessible.
How OpenSmartRoute helps teams route requests to these assets
OpenSmartRoute helps teams manage requests for simulation assets. It sends requests to the best-fit model or agent. It scores candidates on quality, cost, speed, and safety. The team sets weights per request. Hard rules pin requests to specific constraints. A request with personal data stays on-premises. A region or cost cap is never crossed. The router learns from outcomes over time. A model that answers well gets more traffic. A model that fails gets less traffic.
What to do next to find or contribute assets
Teams should find assets on Hugging Face. They can check the SimReady Central library. They can search for equipment types they need. They can download assets for their simulation projects. Teams that make equipment should contribute assets. They can use the submission skill to validate files. They can package and submit assets to the library. This helps grow the open resource for everyone.
Imagine building an AI agent that has to decide whether a customer complaint needs human intervention, which department should handle a query, or whether a submitted delivery photo is valid. Getting these decisions right matters. A false...
